Chat bot with customizable Attention Layer.

Basic features of this code base are:

These are the placeholders for our model’s inputs. Based upon the feature(customizable) selection, these model input’s are consumed by model.

Some of the hyper parameters and they are used for faster training of the model and avoiding over fitting.

Encoder layer will return two values and based upon the customizable feature selection, either of these value is used. In case where we don’t want to use “Attention Layer”, output of encoder last cell i.e. “state” will be used as input by the decoder first cell. In case where we want to use “Attention Layer”, then output of each encoder cell i.e. “output” will be used to predict the decoder output.

Utility method used to concatenate the bi-directional encoder state’s and generate a tuple for decoder layer further processing.

Values of hyperparam, where I’m getting the best results.

Refer above mentioned blog for more detailed on this part.

Adding <PAD> just to make all the sentences of a particular batch of similar size.

Refer above mentioned blog for detailed explanation.

Piece of code is used for predictions.

I have compiled and verified this code, still in case you guys find any issue or have suggestion, please post them.

Related posts:

Making the application better is what we do and fixing bugs as soon as we find them is how we do it. Features also make the application better and usable. This Saturday’s release is a lot bigger then…

Print in Python 2 is a very simple task that can be accomplished using the print statement. The print statement is a function that takes a string or a list of strings as an argument and prints them…

Humanity has awakened to this new decade as a newborn babe ushered into the world with a hard slap on the ass. The very clinical hand of SARS-CoV-19 has roused us out of the warm slumber of high…